> If cosmic rays flip bits in storage or on the network, that can be detected through error coding. But there's no analogy for a CPU that allows cheap online verification of its correctness.<p>Note that CPU can be represented as mostly a set of memory cells, so some techniques for memory correction can probably be used with CPUs as well. <a href="https://bailleux.net/pub/ob-project-gray1.pdf" rel="nofollow">https://bailleux.net/pub/ob-project-gray1.pdf</a>
Are these errors consistent for the same instructions? For example, in the same ALU, will 2+2 always equal 5, or will it spontaneously produce 5 and not happen again for a "long while"?
It's not just CPUs that can cause Silent Data Corruption errors (SDCs). Essentially any chip in the system can give bad results, and those bad results are often not detected.